(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Children's Library | Biodiversity Heritage Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
Upload
See other formats

Full text of "The Linguistic Power in the Analysis of Generalized Descriptions"

THE LINGUISTIC POWER IN THE ANALYSIS 
OF GENERALIZED DESCRIPTIONS 



IN6ER BIERSCHENK^ 
University of Lund 



Running head: Generalized Descriptions 



1. This article was accomplished during the author's visit as 

post-doctoral fellow at the Department of Linguistics, University 
of Ottawa the academic year 1981-82. 



Summa ry . This article presents and discusses a model for analyzing abstract 
invariant relationships between concepts and conceptualizations, linguistically 
expressed in generalized descriptions of information processing. The model 
assumes conceptual information to be directly picked up by means of prepositions 
distinguishing intention from orientation. The precise organizing functions of 
prepositions have led to the construction of an algorithm which makes possible a 
generalized descriptive coding of concepts and a generation of registers with 
functionally defined content. 



Generalized Descriptions 
- 2 - 



We can perceive objects and events, pick up information, and give 
generalized descriptions to our perceptions. This ability has important 
implications for a linguistic study of language. For example, generalized 
descriptive language expressions are assumed to result from our intentional 
abstractive attention and ability to formalize. As a consequence, they have 
definable linguistic forms and are endowed with structures characterized by a 
certain degree of stability. The result of a given abstraction, i.e., a concept 
or a conceptualization, may exist in a variety of linguistic manifestations are 
systematic and appear meaningful. If changes in form take place according to a 
definable pattern, this would mean that generalized descriptions have 
computational linguistic power. Powerful linguistic descriptions make possible 
the construction of an algorithm for predictive purpose, while their explanation 
would require empirical experimentation. Attempting at a rigorous determinism 
of changes in linguistic form is not worthwhile to pursue, however, 
since lexicographers have had minimal success in providing complete definitions 
for such changes. After all, theories based on syntactic models developed by 
linguists and language oriented computer scientists have not been very 
successful in explaining information processing based on human language. 

Formalization of Scientific Observations 

The basic premise on which all scientific observations 

rest is that processes can be isolated and studied with respect to their 
structural stability. Thereby it is implicitely stated that prototypical forms 
are recurrent and observable. This means that they can be formalized, named, 
and communicated. Consequently, communication is a process which becomes 



Generalized Descriptions 
- 3 - 



meaningful only in terms of regularized information processing. The most 
immediate impact of this regularization in formal logical or technical terms is 
probably on the individual's way of thinking. Especially the development of 
advanced communication technology may be expected to have an inevitable 
influence on people's ability to analyze and abstract. It seems above all 
necessary to go beyond a mere acceptance of personal and therefore unexamined 
propositions made by individual computer scientists and to study experimentally 
the impact of highly formalized procedures and conventionalized concepts if one 
wishes to estimate their trivializing effect on available or potentially 
available information. 



Scientific concepts are communicated through scientific documents, whose 
various statements are based on empirical observations about events (Sager, 
1977). The manifest statement of observations may be described as a sentence, 
defined as Noun, - verb - Noun2» where the verb denotes the relation 
between two nouns. A sentence will in the following discussion be referred to 
as a Nj^ V N2* These terms can be used in a description of how semantic 
relations are marked in natural language, a reason why this linguistic formula 
has been considered powerful and useful for a computational approach. This form 
can namely be used to determine an assertion as a predicate - argument 
statement: v(N,. ..). This type of logical formalization is characteristic 
of query languages. By computing natural language on the basis of powerful 
formal systems such as the predicate calculus, it is easy to give simple 
linguistic descriptions of the process because the description is axiomatic and 
deduction with this system is formalized to such a degree that binary coding and 



Generalized Descriptions 
- 4 - 



mechanization become possible. Further, association theory is fundamental for 
the formalization process. 

The association theory assumes that independence exists between two 
linguistic elements and that their similarity is expressible as a logical 
relation between interaction and union of these two elements. A general idea of 
the power in a formalized linguistic expression of this type may be derived from 
the following examples: 



Attribute (Object, Event) = Value 



(1) 



? (Object, Event) = ? 



(2) 



On the basis of example (1) or its modification in example (2) it is possible to 
determine what information can be accessed in a data base containing linguistic 
data. Typically one can ask for generalized descriptions which are 
characterized by kejwords or descriptors. By means of the modified model (2) it 
would also be possible to supply all descriptions that can be characterized by 
specified terms. A common approach to the evaluation of data base systems 
requests an assessment of its performance by some people on the basis of their 
stated profiles of interest. These are usually operationalized by means of 
normalized terms in combination with Boolean algebra. Some other query 
languages and their interactive functioning are described by Lochovsky and 

Tsicheritzis (1981). The results from performance trials are evaluated 
according to a 2 x 2 matrix, where the successful matches of search terms with 
terms in the data base under consideration are represented by the x - axis. 



Generalized Descriptions 
- 5 - 



Thus, a system's performance may be expressed by the relations (B/A+B), which is 
conventionally labelled "retrieval", and (B/A+D) , which has the label 
"precision" attached to it. If one could set up an optimally functionning data 
base system, both relations would express a perfect correlation (1.00), since 
the cells A and D in the matrix would have zero as their value. 

More elaborated efforts in formalizing natural language expressions have 
been undertaken (Quillian, 1968; Winograd, 1972; Anderson, 1973; Schubert, 
Goebel St Cercone, 1979). Common to these approaches is the development of 
different network formalisms. The starting point was Ross Quillian 's memory 
model assuming that human cognition is associatively organized. These so-called 
semantic networks are built out of primitive elements variously called features, 
properties, concepts or primitive propositions, which are said to exist in 
memory. The constructions of semantic networks make clear that this 
representation is founded on the classical-philosophical way of organizing 
knowledge of the world, similar to generic and other lexical-semantic 
relationships. Words are represented as a static collection of elements, and 
elements are related by links defining the restrictions placed on possible 
combinations. Although developing formal models is compatible with some 
"indeterminacy", since deduction is an indeterminate operation, they cannot deal 
with problems of undecidability, e.g. know whether a proposition is or is not a 
consequence of a set of propositions (Thorn, 1975). Thus, it is not enough to 
develop a powerful mechanism for logical deduction, i.e., the application of 
rules of inference to statements made in an axiomatical and formally clarified 
language, whose "semantics" is well specified. It is also necessary to account 



Generalized Descriptions 
- 6 - 



for contextual relations such as they emerge in natural language sentences and 
their dynamic properties. 

Recent trends in network development seem to make the provision of a 
restricted context through a construction of so-called topic hierarchies. These 
have the function to prevent the associations from "exploding" in information 
search procedures. But since the "intelligence" of the computer can reflect 
knowledge only as a verification process, i.e., two or many valued logic, 
dynamic principles cannot be used in information search. Therefore, what is 
accomplished is an automatization of the philosophy and logic built into classic 
organization of information, and unfortunately, the analogy made concerning 
theories of structuring and functioning of human cognition is one main reason 
for the maintenance of traditional reasoning in the organization and set up of 
information processing systems. Recognition, for example, is conceived as an 
attempt ot match input strings or frames with those already stored. Stored 
elements are usually content-addressable and matching is a matter of 
compatibility and composition. As opposed to semantic networks, whose 
organizers have to account for how many "semantic primitives" are needed for a 
syntactic computation of "concepts", a structure based information processing 
approach tries to explore the advantages of taking into account intention and 
orientation of the operating process. This approach utilizes the relations 
between abstractions, which means that information need not and cannot be 
treated as form only, as it is done in the network, but is considered to be the 
outcome of a purposeful process and thus embedded in a structure. 



Generalized Descriptions 
- 7 - 



A Structure Oriented Model 

Of fundamental importance for my discussion in this article is my usage 
of the notion "structure". It refers not only to the process of construction 
but also to the actual components and their relations within a given unity. 
Thus, I do not use the term merely to make reference to stability or generalized 
patterns of linguistic elements. On the other hand it should be clearly 
understood that one cannot point to the structure of a system. It is always 
necessary to state its form of organization. On the basis of this specification 
it should be impossible for anyone to speak of the "structure of English" and 
the "structure of music" or "conceptual patterns", because this would be a 
confusion of the notion "structure" with the notion "generalized patterning". A 
structure may remain stable over long periods of time, or it may change from 
moment to moment, depending on the characteristics of the ongoing cognitive 
processes within that structure. Consequently, the term structure may not be 
confused with the concept of stability. This definition recognizes that 
information and structure are connected and that all changes of information over 
time may be regarded as process. 

When the purpose is to Inform about a sequence of events, as e.g. a 
scientific process, a structure oriented model is required. For Indo-European 
languages the Agent-action-Ob ject paradigm seems to be general. This model is 
dynamic in the sense that it takes intentlonality and orientation into account 
and can therefore be used in cognitive studies of Information processed from 
whole texts. The AaO paradigm has been discussed and defined in Bierschenk and 
Bierschenk (1976, Chapter 2). The components represented by the paradigm should 
not be confused with a case model of the Fillmore type. Fillmore's model is not 
basically different from other philosophical models, since it organizes the 



Generalized Descriptions 
- 8 - 



world mainly in semantic-logical terms. On the other hand, the correspondence 
between the N^^ v N2 and AaO models is evident in the "function 
component". The "a" component is required to identify the parts of an 
assertion. The action denotes which object(s) or goal(s) must be present in 
order for a conceptualization to be detected. But fragments of a concept or a 
conceptualization may also be present in connection with the AaO paradigm, i.e., 
single values may be missing and need to be supplemented. To accomplish this, 
default variables are used. By means of the AaO model latent dimensions in the 
information structure can be detected, and inferences are made to the relations 
holding between concepts and conceptualizations (B. Bierschenk, 1982). 

A model intended to represent a proposition about the scientific work 
communicated in a research report cannot, contrary to the view held by Sager 
(1977, p. 86) be linguistically based, but should be related as closely as 
possible to the cognitive foundations regarded as adequate to the research 
process supposed to be represented by the content of the document. It is 
generally accepted that "problem", "method" and "goal" are the fundamental 
components in the research process (Bunge, 1967, p. 6). The Method component 
explicitly denotes the way in which the search for new information is to be 
done. The "Problem-method-Goal" model denotes the aim (direction) in the 
research process, namely a conscious steering towards or a systematic and 
goal-oriented search for new information. An abstraction of invariants from 
scientific text will in the following discussion be referred to as the PmG 
paradigm. The single components of the paradigm have been presented in B. 
Bierschenk (1974). 



Generalized Descriptions 
- 9 - 



The components of this abstract model should, like abstractions in 
general, be regarded as invariants of concepts and conceptualizations. It may 
be appropriate to mention that the three components are not comparable with such 
linguistic categories as, e.g. word classes, sentence constituents or cases. 
The appearance of a "verb" is readily expected under the Method component. But 
in fact, a research technique (which is one of several realizations of the 
method) may be called "an interview", and a problem (what a researcher tries to 
solve) may be referred to as "to interview" . 

Analyzing a generalized description such as a title assigned to a 
scientific text from the point of view of a linguistic manifestation of 
conceptual relations (here scientific concepts and their relationships) implies 
a study of the result of conceptual information processing and thus an 
abstractive attention to a scientific text as manifested in a title. Since the 
overt organization of the title may be fragmentary and restricted with respect 
to the PmG paradigm, a default variable is required even in this case, serving 
as a place holder for what is missing in the abstraction. In order to analyze 
the relations holding between the single components in the title, a way of 
indicating the components roles and structural connections is needed, referring 
to the theoretical starting points of the model. 

Differential Linguistic Functions 

The basic linguistic elements used to relate concepts and to structure 
reality are prepositions. Depending on the model of language under discussion, 
the prepositions are labelled "morphemes" or "lexemes", where the former applies 



Generalized Descriptions 
- 10 - 



to syntactic and the latter to semantic function. To study cognitive 
development and language acquisition both functions have been employed in 
psycholinguistics and cognitive psychology (Brown, 1973). These kinds of study 
refer mainly to the structural function. The organizational function is 
emphasized in computer applications of text analysis. A discussion of the 
various theories may be found in 1. Bierschenk (1981). Based on this 
discussion, it can be hypothesized that prepositions constitute the intentional 
and orientational functions in both the process of building up a structure and 
in producing a form. This may be illustrated with an example from a 
transformational stage. From an interview about research there could be a 
statement like 



I have analyzed titles for several months 



(3) 



where the researcher is present (I) and tells about a process that has happened 
(verb forms) and also for how long time. When the same author writes a report 
about his research, the time specification will not be present in the title 
(there is no "here and now") and his person is implicit: 



An analysis of titles 



(4) 



The "event" condensed here derives both the agent and the action from An 
analysis . The transformational level is marked by the preposition of . Before 
the transformation the verb form indicated the concept titles as being an 
object. This role is still to be discerned after the transformation through the 
function of the preposition of. 



Generalized Descriptions 
- 11 - 



The goal of scientific inquiry, however, is not to handle persons or 
solid objects, but to deal with problems. Problems also imply that there are 
possible solutions, which means that problems determine the research process in 
the same way as the object in a sentence determines what type of verb may be 
selected. Therefore, it might be justified to have the label "object" replaced 
by the label "problem". Problems include intentions; consequently, the role of 
the component in the PmG paradigm is comparable with the Agent in the AaO 
paradigm. The preposition o£ then functions as an operator for the Problem 
component. When the problem has been identified. An analysis remains to be 
analyzed. This part of the title can now be given an unambiguous 
interpretation, i.e., it denotes the scientific event, as formalized through 
methods or means used. 

Scientific activity cannot be equated with a determinable object or 
determinable problems, but should be defined as a strategy, i.e., a way of 
tackling problems. The development of new methods and instruments increases the 
individual researcher's possibilities of creating new information. For this 
reason, the informative value of the title increases if it contains information 
on the research techniques used. If the information is expanded, so that the 
analytical technique is made explicit, the title can be given as 



An analysis of titles with a coding algorithm 



(5) 



As shown sofar, the research strategy is An anlysis . The preposition with, 
however, specifies in more detail what plans, techniques, or instruments have 
been employed. Means or instruments play a central role in science. It is 



Generalized Descriptions 
- 12 - 



therefore reasonable to assume that with a coding algorithm is an explicit 
expression of what researcher X does, that is, his way of analyzing titles, and 
that this is conceived to be an important piece of information to communicate. 
Therefore all concepts denoting means may be arranged under this Instrument 
component. In research, these are seldom solid objects (tools). However, they 
do have a more concrete function in connection with the method. This is the 
reason why the Instrument component is discussed and placed between Method and 
Goal. The Goal determines the instrumentation of a method. 

If the title is further expanded, so that it also contains an explicit 
statement of the goal, it may appear in the following form: 

An analysis of titles with a coding algorithm for concept recognition (6) 

The goal is in this case to recognize concepts. As the example (6) shows, the 
preposition for denotes this intention, i.e., it gives the reason why a certain 
act that requires certain instruments has been performed. 

Research focuses on problems, which are multifaceted. This implies that 
a scientific title gives expression to multi-dimensional phenomena. But 
problems, too, may be determined as being part of a problem area, which 
incorportas a time dimension. This is expressed in the title: 



An analysis of concepts in titles from four decades 



(7) 



Generalized Descriptions 
- 13 - 



The preposition in_ orients to where the concepts are to be found, I.e., 
in titles and In no other type of text. The preposition ^^ here denotes a 
demarcation In that It specifies the contextual domain of the problem under 
consideration. The titles must not be regarded as expressing a concrete place 
or container In this conceptualization, which Is why a philosophical and thus 
form oriented model of inclusion will not do. The preposition from is used in 
contexts of space denoting a starting point. It has the same orientatlonal 
meaning when used on the time dimension; it demarcates the temporal range. It 
should also be mentioned that geographic places, too, are regarded as abstract 
concepts in this model. 

As can be seen from the above examples, a strict sequential order is 
described. The last demarcation specifies concepts in such a way that it 
concerns not only titles but also a particular period of time. There seems to be 
a need for a time dimension in order to discriminate between experience from 
different periods of time. 

Construction of an Algorithm 

The creation of order presupposes a form or organization within which the 
order is to be set up. Thus the task of creating order within titles first of 
all implies the determination of what constitutes a title. A detailed 
discussion of the order-creating rules and coding procedures are to be found in 
I. Blerschenk (1981). As Indicated in the previous section, the analytical 
method employed rests on certain basic assumptions concerning the function of 
the prepositions in a title, which is to indicate or point foward towards 



Generalized Descriptions 
- 14 - 



certain types of concepts. In this function their position relative to one 
another is of considerable importance. From the model it is obvious that the 
prepositions ("pointers") belonging to the main components are essentially of , 
with and for . These prepositions propel "the action" fowards, thus expressing 
the transitivity in the paradigm ("the horizontal level"). Prepositions 
demarcating the main components are located between them and in a specified 
order. As a consequence, the concepts they point foward towards are ordered 
"vertically under the nearest preceding main preposition. 

Within the theoretical context presented the former type of preposition 
will be called intentional . On the assumption that a concept is perceived as a 
denotation with respect to the characteristics that form the connections of the 
concept, an explicit demarcation should be considered spatial and as such it 
functions as a demarcation, i.e., as a visually signalled orientation. 
Therefore, the latter type of prepositions will be called orientatlonal. 

The sequential order among the prepositions is employed by way of 
numerical codes expressing the relations between intention and orientation 
within a coordinate system as follows: 



An analysis of concepts in titles 
40 30 33 



(8) 



The concepts expressed by the intentional prepositions are assigned numerical 
code numbers ending with "0". The associated orientatlonal concept is assigned 



Generalized Descriptions 
- 15 - 



another number, in which the sequential order begins with a number other than 
zero. The following concept would have the numerical code 34, and so on. The 
Method component is assigned code number 40. This system is set up in such a 
way that the sequential order is algorithmically fixed. For tranparency 
reasons, the coding system has been taken over from the ANACONDA system 
(Bierschenk & Bierschenk, 1976, p. 40). It should be kept in mind that the 
ANACONDA model is applied to natural language (interviews), thus including more 
components than does the PmG model. Certain codes therefore are left empty in 
PmG, others are processing a generalized meaning because of the generalized 
description and thus higher order abstraction in the PmG model. The numeric 
codes for conceptual coding in this study are the following: 



Problem 


30 


and 


Method 


40 




Goal 


70 


and 


Instrument 


80 


and 



Extensions 33, 34, etc. 

Extensions 73, 74, etc. 
Extensions 83, 84, etc. 



Since the system is based on distinguishing main concepts from subconcepts, the 
rules have also been constructed in such a way that titles of considerable 
length will be assigned a correct coding. A more detailed description of them 
together with the computer program are to be found in Bierschenk, and 
Sternerup-Hansson (1979). 



Generalized Descriptions 
- 16 - 



Structural Stability in Scientific Observations 

The goal of testing an analytical method of the kind presented here 
requires a decision regarding who or what should represent a certain field of 
application (here research within education). Based on the results presented in 
B. Bierschenk (1976) the definition of the researcher population employed here 
includes psychologists, educationists and sociologists. After "researcher" had 
been defined, a random sample was drawn from the resulting population. The 
knowledge represented by these researchers (B. Bierschenk, 1979) can be 
recaptured in non-fugitive form from their written works, which have all been 
collected, starting from their first scientific product (Ph. L. or Ph. D. 
Thesis). Thus this sample of works may be regarded as representative of the 
focus of attention in research of relevance to Swedish education. 

Data base . Subjecting the model described to an empirical study cannot be done 
without a data base. Depending on the purpose of an analysis it may be 
organized in different ways and created by means of different techniques. The 
data base to which the method under discussion has been applied is experimental. 
It contains (1) all bibliographic data concerning scientific documents produced 
during a period of 40 years, (2) all references cited by the researchers in the 
documents from this period, and (3) a linkage between the references given in 
the documents and an extensive interview material (4000 typed pages) concerning 
the researchers grant-supported activities. A more detailed account of the data 
base is given in I. Bierschenk (1981). 

Form and Structural Stability . According to the analytical model, the 
conceptualizations may be more or less explicitly stated. In the most explicit 
case they are represented by the components 30+40+(80)+70, together with 



Generalized Descriptions 
- 17 - 



possible attributes. Thus, the structural stability of a pattern implies that 
certain patterns are possible, common, uncommon, or impossible. Further, there 
is an in-built restriction in the system, among other things due to the 
sequential order of the subordinate codes. 

For a quantitative description of patterns, profiles were printed 
containning all the existing types together with frequency counts. They showed 
that there are 85 different patterns, 42 of which (50%) arc unique. The latter 
types are not very suitable for a quantitative description of the material. 
Instead, the focus of interest is on certain recurrent patterns, so as to make 
it possible to discover regularities, which is a prerequisite for the 
development of algorithms for automatic analyses. Therefore it was decided that 
the reference base should be used as a control base. The base of the original 
works of the researchers is to a certain extent a subgroup of the reference 
base, and so the pattern profiles were also counted on the reference base. The 
references represent 241 patterns, 89 of which are unique. The number of 
different patterns in the references is higher than in the work base. At the 
same time, however, the reference patterns are characterized by a larger number 
of common patterns than the works, since only a third of them are unique. A 
comparison between the patterns in the two bases thus makes it possible to 
estimate, with some degree of confidence, the stability of the general 
properties of the work profiles. 

To determine the commonality of patterns in the works and the references, 
a lower limit was needed. The works have been produced by 40 researchers. If 
the same pattern occurs either four times in the works of one person, or 



Generalized Descriptions 
- 18 - 



conversely, once in each of four persons' work, this means a frequency of 10% in 
the sample. Frequencies under 10% may be regarded as random variation. 
Therefore it was decided that only patterns with a frequency of 5 or more should 
be considered in the comparison. 

The patterns of the works were ordered according to their ranks and then 
compared with the ranks of the references for the respective pattern. 
Spearman's rank correlation was calculated and found to be high (rs= .89) (The 
inspection of the material made it evident that editing of the references would 
have resulted in an even higher correlation.) The result is based on some 9,500 
titles, quite a high number in this kind of study. 

Functional Analysis and Differential Power 

On the basis of the result accounted for in more detail in I. Bierschenk 
(1981, pp. 80-90), it appears natural to take a closer look at the main types of 
work produced by the researchers and attempt to determine whether there are 
particular structures which differentiate types of documents. Since this type 
of research covers a wide range of activities, it should be expected that there 
exists an interaction between the structure of a title and the form of 
representation chosen. 

As a basis for the comparison type of pattern is used. The first 
criterion for grouping concerns the two main patterns, namely the difference 
between Problem (30) and Method (40). In this way patterns with and without 40 
are distinguished. Then the patterns are analyzed according to intentionality 
and orientation, indicating structural complexity at different informational 



Generalized Descriptions 
- 19 - 



levels. The 40 type is process-oriented, the 30 type problem-oriented, which 
means that the former relates phenomena whereas the latter describes and 
demarcates one and the same phenomenon. Further differentiation then results 
first in the group 40+30, 40+70, and 40+80, representing explicit intentional 
relations between concepts. The corresponding pattern of the other type is a 
single 30, since the problem orientation is characterized by implicit intention. 
Thus, explicitly stated intentionality forms one main group, and implicit 
intentionality forms the other. The degree of structuralization is not assumed 
to increase with connective relations. This implies that relations such as 
40+30+30 and 30+30+30 within the respective groups are allowed. 

A further distinction is now necessary. It concerns the presence of 
orientation within each main group. Among the patterns in the first group it 
can be seen that only the 40+30 type is followed by extensions. If the 
connectivity rule is to be followed (as it should be), the pattern 40+30+33 is 
also formed. As a consequence the pattern 30+33 is given within the other main 
group. 

If the degree of complexity grows, one more pattern in each main group 
can be formed, namely 40+30+33+34 and 30+33+34 respectively. 

Interaction of Main Types of Form with Structural Variation 



The structural stability of the functional operators will now be analyzed 
by means of a comparison between document types and pattern type previously 
discussed. Eight groups could be discerned. They were ranked according to the 



Generalized Descriptions 
- 20 - 



frequency of the entire group and tabulated against twelve document types as 
shown in I. Bierschenk (1981, p. 92). The results of this tabulation may be 
presented in the following way. The pattern type 30 cannot be said to be 
typical of any particular form of representation. The fact that titles can 
display more than one pattern may be the reason why this pattern type 
functionally does not differentiate between types of representation. The type 
30+33 is very frequent, too. A problem component, single or together with one 
extension at most appears in all representations, and so does not function as a 
typical pattern representing a certain sample of document. 

When it comes to 40+30 it can be considered a typical pattern. It is the 
most common pattern in generalized descriptions which are headings of chapters 
in a book edited by someone other than the author. The pattern 40+80 is typical 
of monographs, whereas textbooks are of the 40+70 type. The monographs in this 
material include some theses whose basic property is the explicit statement of 
instruments and techniques used for the testing of a method. The textbooks 
investigated state explicitly for whom or what they are intended, that is, the 
goal may be certain groups of persons, a grade in school, etc. 

The 40+30+33 pattern is typical of research reports. Thus reports have, 
as opposed to the others, both intention and orientation explicitly stated. A 
higher degree of orientation is in general expressed in the titles of scientific 
journal articles, but the type pattern is 30+30+34, thus there is 
no explicit expression of intentionality. The interaction between structure and 
type thus emerging may be summarized as follows: 



Generalized Descriptions 
- 21 - 



(40+70) -^[Textbooks ] 
(40+80) —♦•^Monographs ] 
(40+30) -:a«» (chapters in booksj 
(30+33+34) ->{ Journal articles^ 
(40+30+33) ->{Research reports^ 

The structural variation in the generalized descriptions thus far analyzed seems 
to have a differential function with respect to the form of representation 
chosen. However, without detailed experimental studies it is difficult to 
determine the extent to which the conceptual structures discovered with respect 
to the particular groups correspond to the information they are intended to 
communicate. 

Linguistic Form as Carrier of Conceptual Information 

In this section it will be demonstrated how the coding algorithm has 
worked out on authentic material. Apart from some titles (about .005%) with a 
low degree of abstraction compared with expectancies and structural logic built 
into the mechanism, the conceptual decoding has resulted in some data registers 
or files corresponding to the various components of the model. These files are 
now functionally related to each other because of the non-philosophical 
classification. The structural principle underlying the model presented in this 
article also reveals such dimensions that a manual analysis with a classical 
orientation towards philosophical classification schemes could have performed 
only with difficulty. To illustrate the difference I would like to discuss the 
title 



Generalized Descriptions 
- 22 - 



Integration of children with handicaps 
MP I 



(9) 



According to the model children is the problem dealt with in this study and the 
methods and techniques are abstracted in the concept integration ; the different 
steps to take are not explicit. A linguistic analysis, when the model of 
interpretation is anchored in natural language variation, would regard the with- 
phrase, i.e., the concept handicaps as associated to children, which classifies 
it as a property. The research-oriented information model, however, assumes 
this property to be instrumental. In reality, this should be the reason why the 
author expresses or explicates the instrument. It is likely that the 
methodology of integration is different for those children. Therefore, it was 
of special interest in this report. Thus integration may be a practical way of 
handling the children and also a method of study. Examples of variability of 
methods generated are "research" (incorporating several actions), "reflections" 
(a way of reporting one's result), "hand-book" (a kind of methodological 
strategy in educating research students). Consider also 



Goals for teacher training 

M G 
Goals in teacher training 

P 



(10) 



(11) 



Generalized Descriptions 
- 23 - 



The concept goals is a method in the first case, representing activities among 
these researchers involving goal description. It has a clear methodological 
meaning within educational technology, in which teacher training is the overall 
goal. The preposition for recognizes the first goal (10) as a method, whereas 
the preposition in codes it as a problem, since an explicit intentionality does 
not exist in the second case. Thus goals in teacher training just specifies the 
context within which a certain problem is dealt with. That goals is a noun is 
not of any import in the functional oriented registers. The author may discuss 
the same goals in the two titles, but from different view points, from different 
functional domains. This also makes the following title functionally 
communicative to the information searcher: 



School for the 80 's 



(12) 



It was written 20 years ago when the 80 's still begonged to the future. The 
method component is here given a broader meaning, since the school may also be 
seen as an instrument. Method and instrument are components which can form 
method- (instrument)-goal hierarchies in relation to the degree of complexity in 
the desired goals. To reduce method and instrument to a simple concept the term 
"means" is used. In the light of the theoretical assumption and knowledge of 
this author's activities and field of inquiry in Swedish educational research, I 
believe that the proposed interpretation can be validated. 



Generalized Descriptions 
- 24 - 



Conclusion 



In the analysis two kinds of prepositions have been distinguished, 
prepositions referring to intention, and prepositions referring to orientation. 
These have precise organizing functions which have led to the construction of an 
algorithm, which makes possible a conceptual coding of generalized descriptions 
and the generation of registers with functionally defined content. 

An analysis of the structure of titles shows that patterns can be 
detected that are typical of certain types of form and less typical of certain 
other types of presenting scientific information. The experimental results show 
that the most characteristic aspect of the science field investigated is the 
Problem component, single or with one extension. Because of the intentionality 
inherent in the model, this pattern should be interpreted as expressing an 
implicit intentionality. If two components appear, they are most likely a 
Problem and a Method. The presence of the method should be Interpreted as an 
explicit expression of intentionality, as in research itself. The more complex 
the patterns are the less often they appear. There also seems to be a tendency 
that many extensions prevent "transitions" between components. 

The informational value of this conceptual structuring has been analyzed 
by covariating the pattern types with types of document. For example, textbooks 
and other kinds of monographs are less complex than Journal articles, which in 
turn are less complex than research reports. Thus three general structures 
appear in the material, namely (1) explicit intentionality, (2) implicit 
intentionality + second degree extensionality, and (3) explicit intentionality + 
first degree extentionallty. 



Generalized Descriptions 
- 25 - 



Finally, the present analysis may be regarded as an initial attempt to 
develop methods and techniques for a systematic analysis of language as regards 
what information is to be conveyed, whereby questions dealing with the structure 
assumed to characterize a message receive special emphasis. 



Generalized Descriptions 
- 26 - 



REFERENCES 



Anderson, J.R. Computer simulation of a language acquisition system: A first 
report. Jn R.L. Solso (Ed.) Information processing and cognition: The 
Loyola Sjnnposium. Hillsdale, N.J. : Erlbaum, 1975. Pp. 295-349. 

Bierschenk, B. Perception, strukturering ach precisering av pedagogiska och 
psykologiska forsknlngsproblem p3 pedagogiska instltutioner i Sverige./ 
Perception, structuring and definition of educational and psychological 
research problems on departments of educations research in Sweden./ 
Pedagogisk-psykologiska problem . No. 254, 1974. / In Swedish/ 

Bierschenk, B. En longitudinell analys av kunskapsutvecklingen inom 

utbildningsforskningen. / A londitudinal analysis of knowledge development 
within educations research. Pedagogisk-psykologiska problem . No. 355, 
1979. /In Swedish/ 

Bierschenk, B. An ecological model for the processing of symbolic information. 
Perceptual and Motor Skills , 1982, 54, 663-674. 

Bierschenk, B. & Bierschenk, I. A system for a computer-based content analysis 
of interview data. (Studia Psychologica et Paedagogica , 32.) Lund: 



Gleerup, 1976. 

Bierschenk, B., Bierschenk, I. & Sternerup-Hansson, A. Ett datorprogram for 
syntaktisk kodning av titlar till vetenskapliga skrifter. / A computer 
program for syntactic coding of titles of scientific documents./ 
Testkonstruktion och testdata . No. 34, 1978. /In Swedish/ 

Bierschenk, I. An information processing experiment. A method for the generation 
of an intermediate language for representing scientific information. 
Didakometry , No. 62, 1982. (Doctoral thesis, Dept. of Computational 
Linguistics, University of Gb'teborg, Sweden.) 

Brown, R. A first language. The early stages . Cambridge: Harvard University 
Press, 1973. 

Bunge, M. Scientific research I. The search for system . New York: Springer, 
1967. 



Lochovsky, F.H. & Tsicheritzis, D.C. Interactive query language for external 
data bases. Toronto: Computer Systems Research Group, University of 
Toronto, 1981. 



Generalized Descriptions 
- 27 - 



Qulllian, R. Semantic Memory. In M. Minsky (Ed.)> Semantic information 
processing . Cambridge: MIT Press, 1968. Pp. 216-270. 

Sager, N. Perspective paper: Computational Linguistics. In D. Walker, H. 
Karlgren, & M. Kay (Eds.), Natural language in information science. 
Stockholm: Skriptor, 1977. Pp. 75-100. 

Schubert, L.K., Goebel, G.R. , & Cercone, N.J. The structure and organization of 
a semantic net for comprehension and inference. In N.V. Findler (Ed.), 
Associative networks . New York: Academic Press, 1979. Pp. 721-775. 

Thom , T . Structural Stability and morphogenesis. An outline of a general 
theory of models . Reading, Mass.: Benjamin, 1975. 

Winograd, T. Understanding natural language. New York: Academic Press, 1972.